1 Published online in Population Ecology
2 http://link.springer.com/article/10.1007/s10144-012-0350-5
3 DOI 10.1007/s10144-012-0350-5
4
5 Spatial segregation among age classes in cave salamanders: habitat selection or
6 social interactions?
7
8
9
10 Gentile Francesco Ficetola 1,2 *, Roberta Pennati 2, Raoul Manenti 2
11
12 1: Dipartimento di Scienze dell’Ambiente e del Territorio, Università di Milano-Bicocca. Piazza
13 della Scienza 1, 20126 Milano, Italy
14 2: Dipartimento di Bioscienze, Università degli Studi di Milano. Via Celoria 26, 20133 Milano Italy
15
16 *Corresponding author. E-mail: [email protected]
17
1 18 Abstract
19
20 Within species, individuals with different sexes, morphs and age classes often show spatial
21 segregation. Both habitat selection and social processes have been proposed to explain intraspecific
22 spatial segretation, but their relative importance is difficult to assess. We investigated spatial
23 segregation between age classes in the cave salamander Hydromantes (Speleomantes ) strinatii , and
24 used a hypothetico-deductive approach to evaluate whether social or ecological processes explain
25 segregation pattern. We recorded the location and age class of salamanders along multiple caves;
26 we measured multiple microhabitat features of different sectors of caves that may determine
27 salamander distribution. We assessed age-class segregation, and used generalized mixed models
28 and an information-theoretic framework, to test if segregation is explained by social processes or by
29 differences in habitat selection. We found significant age-class segregation, juveniles living in more
30 external cave sectors than adults. Multiple environmental features varied along caves. Juveniles and
31 adults showed contrasting habitat selection patterns: juveniles were associated with sectors having
32 high invertebrate abundance, while adults were associated with scarce invertebrates and low
33 temperature. When the effect of environmental features was taken into account, the relationship
34 between juveniles and adults was non negative. This suggests that different habitat preferences,
35 related to distinct risk-taking strategies of age classes, can explain the spatial segregation. Juveniles
36 require more food and select more external sectors, even if they may be risky. Conversely, adults
37 may trade off food availability in favour of safe areas with stable micro-climate.
38
39 Keywords a priori inference · predation risk · spatial pattern · spider abundance · trade-off
2 40 Introduction
41
42 The identification of processes determining the distribution of organisms is a major challenge of
43 spatial ecology. Patterns of spatial segregation, defined here as differences in spatial organization
44 among individuals, can occur at both the interspecific and the intraspecific level. At the interspecific
45 level, the segregation at fine spatial scale allows the coexistence of species occupying similar
46 niches, and can be an important driver of biodiversity patterns (Firth and Crowe 2010; Darmon et
47 al. 2012). However, spatial segregation can also occur at the intraspecific level among morphs,
48 sexes or age classes (Formica et al. 2004; Field et al. 2005; Ruckstuhl 2007; Main 2008; van Toor
49 et al. 2011; Bjorneraas et al. 2012). For instance, in several vertebrates, sexes segregate in groups
50 showing distinct spatial organization and resource exploitation (Ruckstuhl 2007; Main 2008). Two
51 types of processes have been proposed to explain intra-specific spatial segregation patterns: the
52 social and the habitat segregation hypotheses. The social segregation hypotheses (SSH) suggest that
53 social processes (selection for neighbours with similar features, intraspecific competition,
54 cannibalism) are the major drivers of segregation. In contrast, the habitat segregation hypotheses
55 (HSH) suggest that segregation is mostly driven by habitat-selection processes, such as the selection
56 of sites on the basis of foraging quality or risk of predation (Ruckstuhl 2007; Main 2008). Under
57 habitat-selection processes, social segregation may occur as a simple by-product of differences in
58 habitat selection (Bowyer et al. 2002).
59 In organisms with complex life cycles, such as many insects and amphibians, the age classes
60 correspond to discrete, morphologically distinct phases that often exploit distinct habitats. It has
61 been proposed that the evolution of complex life cycles can be explained ecologically by the
62 reduction of intra-specific competition (Istock 1966; Wilbur 1980; Moran 1994), with mechanisms
63 somehow analogous to those proposed to explain spatial segregation. Complex life cycles are
64 extremely successful life-history strategies, suggesting that organisms minimizing interactions
65 between life-history stages (which usually correspond to age classes) have strong advantages
3 66 (Wilbur 1980; Moran 1994). The similarity between this ecological hypothesis for the evolution of
67 complex life cycles, and the HSH, suggests that spatial segregation among age classes (hereafter:
68 age-class segregation) may widely occur also in organisms without complex life cycles. However,
69 most analyses on intraspecific segregation have focused on sexual segregation in a few groups of
70 vertebrates (Ruckstuhl 2007; Main 2008). Very few studies investigated the causes of age-class
71 segregation in organisms with direct development (see Cransac et al. 1998; Bon et al. 2001;
72 Ruckstuhl and Festa-Bianchet 2001 for analyses with ungulates).
73 The identification of processes determining spatial segregation is challenging: for example,
74 multiple explanations have been suggested after the observation of sexual segregation in the same
75 species of ungulates (Ruckstuhl 2007). The HSH has been proposed as the most likely explanation
76 of sexual segregation (Main 2008; but see Singh et al. 2010). The few studies that assessed the
77 factors determining age-class segregation hypothesized both ecological differences (Labée-Lund et
78 al. 1993; Field et al. 2005) and aggressive / social interactions (Cransac et al. 1998; Salvidio and
79 Pastorino 2002; Galvan 2004; Harvey et al. 2008), but we are not aware of explicit tests to
80 distinguish the hypotheses. The hypothetico-deductive reasoning is an emerging approach to infer
81 processes from distribution patterns (McIntire and Fajardo 2009), and may help to identify the
82 causes of spatial segregation. This requires well distinct a priori hypotheses, formulated on the
83 basis of biological knowledge (Ruckstuhl 2007; McIntire and Fajardo 2009; Dochtermann and
84 Jenkins 2011), and the application of information-theoretic statistical models, explicitly testing the
85 support of alternative hypotheses on causal processes (McIntire and Fajardo 2009; Ficetola et al.
86 2010; Dochtermann and Jenkins 2011; Symonds and Moussalli 2011). Spatial segregation
87 determines clear distribution patterns of individuals, and the different underlying processes (i.e.,
88 SSH vs. HSH) are expected to produce distinct patterns. If segregation is mostly determined by
89 social mechanisms (SSH), we predict a negative relationship between the distribution of different
90 classes of individuals (age, sexes, or morphs; Fig. 1a) but, when these social relationships are taken
91 into account, classes should show similar habitat selection patterns (Fig. 1b). Conversely, if
4 92 segregation is mostly determined by ecological differences (HSH), we predict different habitat
93 selection patterns between classes (Fig. 1d) and, when environmental features are controlled for, no
94 negative relationships between them (Fig. 1c).
95 European cave salamanders (genus Hydromantes , subgenus Speleomantes ) are ideal
96 organisms for the study of spatial segregation. Cave salamanders have direct development. They are
97 not obligate cave-dwellers but, when external conditions would be too harsh (e.g., dry, hot, and
98 particularly during warm seasons) for lungless terrestrial salamanders, they retreat to underground
99 environments where they find a more suitable microclimate (Cimmaruta et al. 1999; Camp and
100 Jensen 2007; Vignoli et al. 2008; Ficetola et al. 2012). When in the cave environment, they are
101 easily detectable (Ficetola et al. 2012) and show extremely limited displacements; for instance, in
102 H. strinatii , each individual usually occupies areas ≤ 8 m 2 (Salvidio et al. 1994). Therefore,
103 observed distribution patterns reflect the actual occupancy of the cave environment. Furthermore,
104 cave environments have relatively simple habitat features that can be well described by a limited
105 number of parameters representing abiotic and biotic features.
106 Previous studies observed spatial age-class segregation in cave salamanders, with juveniles
107 living more close to the cave entrance (Salvidio and Pastorino 2002). It has been proposed that
108 segregation may occur because juveniles avoid cannibalism, or are losers during competition for the
109 best home ranges (Salvidio and Pastorino 2002), in accordance with the SSH (avoidance of
110 aggression hypothesis; Ruckstuhl 2007). However, the processes determining segregation have not
111 been assessed. Here, we analyzed spatial segregation among age classes in the cave salamander,
112 Hydromantes (Speleomantes ) strinatii , and tested whether the distribution pattern supports the
113 predictions of the SSH or of the HSH (Fig. 1).
114
115 Methods
116
117 Study species and area
5 118
119 Hydromantes strinatii is a small (up to 13 cm) plethodontid cave salamander endemic of a small
120 area between NW Italy and SE France. In Mediterranean regions it can be active throughout the
121 year: it is a hygrophilic species, and can be found in epigeous environments from autumn-to early
122 spring, when outdoor conditions are cool and wet. Conversely, in late spring and summer it usually
123 inhabits caves, crevices and other cavities (Bologna and Salvidio 2006; Lanza et al. 2006). Our
124 study focused on age-class segregation within caves, therefore we performed surveys during early
125 summer, when the importance of underground environments is maximum. We surveyed the natural
126 cavities in two nearby small valleys of Western Liguria: Valle Perti and Valle Ponci (approximately
127 42.20°N, 8.35°E). Additional details on the study area are provided elsewhere (Ficetola et al. 2012).
128
129 Surveys and environmental features
130
131 We performed preliminary surveys in > 30 cavities of the study area, to identify those with presence
132 of cave salamanders. On the basis of these surveys, we selected 11 natural caves with presence of
133 salamanders for more in-depth surveys. We used visual encounter surveys (Crump and Scott 1994)
134 to assess the distribution of salamanders. Each cave was surveyed in June 2011. To minimize
135 variation in outdoor conditions, caves were visited in consecutive days between 11.00 am and 3.00
136 pm; all days had similar temperature and were sunny and dry, as it is characteristic of
137 Mediterranean summer. In each survey, up to six trained people actively searched salamanders over
138 the floor and all the walls of the cave; we explored the caves as deeply as possible, compatibly with
139 our equipment. Previous analyses showed that using these approaches the per-visit detection
140 probability of H. strinatii is high (91%), thus surveys allow a reliable assessment of the distribution
141 of the species (Ficetola et al. 2012).We recorded the location (distance from the cave entrance and
142 height above the cave floor) and measured total length of all salamanders observed. We identified
143 age classes on the basis of secondary sexual characters and size: salamanders above 58 mm or with
6 144 male sexual characters (mature males have mental glands and premaxillary teethes) were considered
145 adults; the remaining individuals were considered juveniles (Salvidio 1993; Lanza et al. 2006).
146 Each cave was subdivided in 3-m longitudinal intervals (hereafter: sectors) for the
147 measurement of environmental features. The size of sectors approximately corresponds to the usual
148 size of home ranges of H. strinatii (4-8 m 2) (Salvidio et al. 1994), even if larger home ranges are
149 possible (Pastorelli et al. 2005; Lanza et al. 2006). The number of sectors in a cave depended on
150 cave size and distribution of salamanders. If at least one salamander was detected in the deepest
151 sectors of the cave, sectors covered the whole cavity. If salamanders were not detected at the end of
152 the cave, sectors covered the whole cave until the position of the last salamander, plus one
153 additional, deeper 3-m sector. For instance, if the deepest salamander in a cave was at 23 m from
154 the entrance, and the cave was deeper than 24 m, we considered nine 3-m sectors for that cave. We
155 recorded the number of adults and juveniles per each sector. Furthermore, in each sector we
156 measured four environmental variables that are known to be the major determinants of salamander
157 distribution in underground environments (Briggler and Prather 2006; Camp and Jensen 2007;
158 Vignoli et al. 2008; Ficetola et al. 2012). Three parameters represented abiotic conditions: air
159 temperature (°C), relative humidity (%) and illuminance (i.e., intensity of incident light, measured
160 in lux). These parameters were measured using a EM882 multi-function thermo-hygrometer and
161 light-meter (PCE Instruments). The minimum illuminance recordable by the light-meter was 0.01
162 lux. Furthermore, as biotic parameter, we counted the number of adult Meta menardi spiders in each
163 sector. In the study area, M. menardi is the most abundant large spider living in the twilight zone of
164 caves (Ficetola et al. 2012; Manenti et al. unpublished data). Meta menardi spiders can be predated
165 by Hydromantes salamanders (Lanza et al. 2006). Furthermore, both salamanders and spiders are
166 predators of arthropods found in underground chambers (Salvidio et al. 1994; Smithers 2005;
167 Vignoli et al. 2006; Novak et al. 2010), and spiders are likely positively related to the overall
168 abundance of invertebrates. Therefore, we considered spider abundance as a measure of the
169 abundance of invertebrates (Ficetola et al. 2012).
7 170
171 Statistical analyses
172
173 We used the “sexual segregation and aggregation statistic” (SSAS, Bonenfant et al. 2007) to test for
174 spatial segregation among age classes. The SSAS has been developed to assess patterns of
175 segregation and aggregation between sex classes; it is based on well known chi-square statistics and
176 its approach is general and can be applied to investigate segregation also in other situations. SSAS
177 provides an estimate of the distance between the observed and the expected distributions of age
178 classes, under the null hypothesis of independence of the distributions of age classes among the
179 groups. Segregation occurs when the ratio n adults : n juveniles in each group deviates strongly
180 from the ratio observed in the whole population (i.e., in all the considered groups). Similarly,
181 aggregation occurs when the ratio in each group is close to the ratio of the population (Bonenfant et
182 al. 2007; Singh et al. 2010). In our study, we considered all the individuals occurring in the same 3-
183 m sector as a “group”. We used a randomization procedure (50 000 replicates) to identify the 95%
184 confidence intervals of SSAS expected under random association between age classes. SSAS varies
185 between 0 (no segregation) and 1 (complete segregation), nevertheless the SSAS values can not be
186 considered as an absolute measure of segregation, and are meaningful only if compared to
187 confidence intervals of expected SSAS (Bonenfant et al. 2007).
188 We used generalized linear mixed models (GLMMs; normal error distribution) to test
189 whether the position of each observed salamander within the caves (distance from the entrance,
190 height above the cave floor) was different between adults and juveniles, and to test whether the
191 environmental variables recorded in each cave sectors were affected by the distance of the sector
192 from the entrance of the cavity. In all GLMMs, we included cave identity as a random factor.
193 Subsequently, we used GLMMs within an information-theoretic approach, to assess whether
194 segregation was most likely explained by the SSH or by the HSH. First, we built a model
195 representing the SSH (social model), relating the abundance of juveniles to the abundance of adults.
8 196 Second, we built a model representing the HSH (habitat model), relating the abundance of juveniles
197 to the four environmental variables. We used Akaike’s information criterion (AIC) to assess the
198 relative support of the two models. AIC trades-off explanatory power vs. number of predictors;
199 parsimonious models explaining more variation have the lowest AIC values. We considered the
200 model with the lowest AIC as the “best AIC” model, and calculated ∆AIC, which is the AIC
201 difference between the models. We then calculated the evidence ratio of the models, which provides
202 a measure of the relative likelihood of one hypothesis vs. another (Burnham and Anderson 1998;
203 Lukacs et al. 2007; Symonds and Moussalli 2011). Using AIC corrected for small sample size
204 instead than AIC would lead to identical results. We also built a model combining social and
205 environmental independent variables, to asses the possibility of a joint effect of environmental and
206 social factors. We then repeated the same procedure for adults, considering the abundance of adults
207 as dependent variable and the abundance of juveniles as the social independent variable. In these
208 models, the dependent was a count (number of individuals), therefore we used a Poisson error
209 distribution. In all models we also reported significance values of independent variables, to
210 facilitate interpretation of their role (Stephens et al. 2007). Estimation of fit of the GLMMs used
211 maximum likelihood; we used likelihood ratio to assess the significance of variables in Poisson
212 models. In all the best AIC GLMMs, residual deviance was similar to residual degrees of freedom,
213 indicating that our models were not affected by overdispersion.
214 In a large number of sectors we detected zero salamanders. To ensure that our results were
215 not affected by zero-inflation, we repeated our analyses three times. First, we run the analyses by
216 excluding sectors where zero salamanders (either adults and juveniles) were present, and by
217 excluding sectors with zero or one salamander. Furthermore, we re-run our analyses using Bayesian
218 Markov Chain Monte Carlo GLMMs (MCMCglmm) (Hadfield 2010). With MCMCglmm, we built
219 the models corresponding to the different hypotheses using zero-inflated Poisson models. We then
220 used the deviance information criterion (DIC) to compare the zero-inflated models with Bayesian
221 MCMCglmms assuming Poisson error (i.e., the same error distribution used in standard GLMMs).
9 222 DIC is a parameter estimating the performance of a Bayesian model on the basis of deviance and of
223 the effective number of parameters; it gives particular emphasis to the random effects. Models
224 having lower deviance and less effective parameters show lower DIC, and are considered to be the
225 models most appropriate to the dataset (Spiegelhalter et al. 2002; Spiegelhalter et al. 2008). For
226 MCMCglmm, we performed 30 000 iterations as burn-in, followed by 300 000 runs with a thinning
227 interval of 100.
228 If needed, variables were transformed using logarithms (distance from cave entrance,
229 illuminance). Correlation between pairs of environmental variables was weak (in all pairwise
230 correlations, | r| < 0.3), indicating that multicollinearity did not pose problems to our models.
231 Analyses were performed using packages LME 4, NLME (Pinheiro and Bates 2000; Bates and
232 Maechler 2010; Pinheiro et al. 2010) and MCMC GLMM (Hadfield 2010) under the R statistical
233 environment (R Development Core Team 2010)
234
235 Results
236
237 Pattern of spatial segregation
238
239 We detected salamanders in all the 11 caves considered. Adults and juveniles cohabited in seven
240 caves; in two caves we detected adults only, in two caves we detected juveniles only. Adults were
241 found at distances of 3-49 m from the cave entrance, while juveniles were found at distances of 1-
242 12 m (Fig. 2). The SSAS showed significant segregation of age classes among the sectors (observed
243 SSAS = 0.048, 95% CI of expected SSAS = 0.021-0.042). Overall, juveniles were detected closer to
244 the cave entrance than adults (mixed model, F1,64 = 31.0, P < 0.0001). Conversely, height from the
245 cave floor was not different between adults and juveniles ( F1,64 = 0.19, P = 0.67).
246 Most environmental features significantly varied with the distance from the cave entrance:
247 in sectors far from the cave entrance humidity was higher ( B ± SE = 1.7 ± 0.6; F1,49 = 7.71, P =
10 248 0.008), while abundance of spiders and illuminance were lower ( B = -0.3 ± 0.1 P = 0.004 and B = -
249 0.10 ± 0.04, P = 0.009, respectively). The relationship between distance and temperature was not
250 significant ( B = -0.05 ± 0.12; F1,49 = 0.16, P = 0.690).
251
252 Distribution of juveniles
253
254 The abundance of juveniles across sectors was not related to the abundance of adults (Table 1a).
255 However, this is not in contrast with the SSAS analysis: this result probably occurred because of the
256 presence of sectors in which the abundance of both adults and juveniles was very low. If sectors
257 with zero or only one individual (either adults or juveniles) were excluded from the analysis, the
2 258 abundance of juveniles was negatively related to the abundance of adults (B = -0.25, χ 1 = 4.1, P =
259 0.041).
260 The habitat model, explaining the distribution of juveniles on the basis of environmental
261 features, had a lower AIC than the social model (Table 1b; ∆AIC = 8.8). On the basis of evidence
262 ratio, the habitat model was 77 times more likely to be the best model than the social model. This
263 model suggested association between juveniles and cave sectors with high spider abundance (Table
264 1b, Fig. 3a).
265 A combined model, considering both environmental variables and presence of adults,
266 showed a lower AIC than the habitat model. However, contrary to the expectations of the SSH, the
267 relationship between the abundance of juveniles and adults was positive, when taking into account
268 the relationship between the abundance of juveniles and the environmental variables. Also in this
269 model, juveniles were strongly associated with sectors with abundant spiders (Table 1).
270
271 Distribution of adults
272
11 273 The abundance of adults across sectors was not related to the abundance of juveniles (Table 1e).
274 The habitat model, explaining the distribution of adults on the base of environmental features,
275 showed lower AIC than the social model (Table 1f; ∆AIC = 11.3). On the basis of evidence ratio,
276 the habitat model was >270 times more likely to be the best model than the social model. This
277 model suggested association between adults and the sectors with less spiders and with cold
278 temperature and limited light (Table 1f). It is remarkable that the relationship between the
279 abundance of adults, spiders and temperature was opposite to the relationship between juveniles and
280 the same variables (Table 1; Fig. 3b).
281 A combined model, considering both environmental variables and the presence of juveniles,
282 showed a lower AIC than the habitat model but, contrary to the expectations of the SSH, the
283 relationship between the abundance of juveniles and adults was positive. Also in this model, adults
284 were associated with cold sectors without spiders (Table 1).
285
286 Analyses using alternative models.
287
288 If the sectors with zero salamanders (either adults or juveniles) are excluded from analyses, the
289 habitat model showed much higher support than the social model (Table 2), confirming the results
290 of previous analyses. For juveniles, evidence ratios suggested that the habitat model was about 90
291 times more likely than the social model; for adults, the habitat model was about 105 times more
292 likely tha the social model. In this case the combined models showed lower support than the habitat
293 model. Overall, the coefficients of the best models were in agreement with the analysis considering
294 all sectors (Table 1, Table 2c). The habitat model is the best model even if sectors with zero or one
295 salamander are excluded. In this analysis, AIC differences among models are limited, probably
296 because of small sample size (Table 2b).
297 For all candidate models, the zero-inflated MCMCglmms had higher DIC than the
298 respective Poisson models (Table 3), suggesting that our results are not affected by zero-inflation.
12 299 For both juveniles and adults, the MCMCglmm with lowest DIC was nearly identical to the best-
300 AIC GLMMs (Table 1d, h).
301
302 Discussion
303
304 When spatial segregation occurs, differences of distribution patterns between categories of
305 individuals may be striking. In our study case, we observed strong age-class segregation in the cave
306 salamander: juveniles were close to the cave entrance, while adults were in deepest sectors of caves
307 (Fig. 1). Multiple environmental features of caves strongly varied with the distance from the
308 entrance. The combination of a priori hypotheses with an information-theoretic approach allowed
309 to assess the relative likelihood of processes potentially determining segregation of age classes
310 (habitat suitability vs. social interaction), and suggested that segregation was most likely caused by
311 different habitat selection patterns (HSH). First, the models assuming that the distribution of age
312 classes is mostly affected by ecological features (HSH; models b and f in Table 1) have much
313 stronger support by the data than those assuming that the distribution of one age class is affected by
314 the distribution of the other class (SSH; social models a and e in Table 1). Furthermore, if the
315 distribution of the other age class is added to the HSH model, the relationship between age classes
316 become positive (Table 1). This pattern is opposite to the expectations under the SSH, which would
317 predict negative relationships between age classes (Fig. 1a).
318 The relationship between the abundance of juveniles and adults was positive, when
319 differences in habitat selection are taken into account (Table 1c, d). This result corresponds to the
320 converse of the prediction of the SSH, yet it is a pattern non predicted by the HSH (Fig. 1). This
321 positive association might occur because both age classes are affected in a similar way by some
322 unmeasured environmental feature. This hypothesis is supported by the analysis excluding sectors
323 without salamanders (Table 2). If sectors without salamanders were not considered, the HSH
324 remains the most supported by the data, but the combined model showed higher AIC than the
13 325 habitat model. The results of the model considering all sectors (Table 1) are probably affected by
326 certain areas having environmental features unsuitable for both adults and juveniles. For instance,
327 both age classes may be very scarce in cave sectors characterized by walls with limited
328 heterogeneity (Camp and Jensen 2007). Finally, we can not rule out the occurrence of unknown
329 social process, determining some positive response to the presence of conspecifics (Gautier et al.
330 2004); additional studies are required to test these hypotheses.
331 Age classes had a contrasting response to major environmental gradients: juveniles were
332 associated with the cave sectors inhabited by abundant spiders, while adults were associated with
333 scarce spiders and lower temperature (Table 1, Table 2, Fig. 3). Meta spiders are among the major
334 predators of arthropods in caves, therefore we considered spider abundance as a proxy of the
335 abundance of invertebrates (Smithers 2005; Novak et al. 2010; Ficetola et al. 2012). Cave food
336 webs mostly depend on external inputs: sectors close to the surface receive more inputs and usually
337 host a richer invertebrate fauna (Hills et al. 2008; Culver and Pipan 2009; Novak et al. 2010;
338 Manenti et al. 2011; Schneider et al. 2011). During summer, both juveniles and adult salamanders
339 feed within caves, and the small home range size (about 4-8 m 2) suggests that they feed
340 approximately in the sectors where they have been observed (Salvidio et al. 1994; Lanza et al.
341 2006). For salamanders, selecting shallow sectors may have the advantage of a higher food
342 availability. On the other hand, inhabiting shallow sectors may determine multiple disadvantages,
343 including micro-climatic conditions more similar to the external ones (e.g., lower humidity) and a
344 highest risk of predation either from invertebrates or from vertebrates entering in the first meters of
345 caves (Culver and Pipan 2009). Actually, some large cave invertebrates (e.g., large spiders) have
346 been observed preying on Hydromantes juveniles (Lanza et al. 2006; Pastorelli and Laghi 2007).
347 Furthermore, the most external sectors often had humidity < 75%, and this may pose physiological
348 problems to salamanders. Humidity influences dehydration rate: inhabiting more humid sectors can
349 allow better maintaining water balance. Plethodontid cave salamanders are lungless and skin
350 humidity / dehydration is also related to respiratory exchanges (Spotila 1972; Briggler and Prather
14 351 2006; Hillman et al. 2009). Conversely, selecting the deepest, coldest sectors would reduce
352 predation risk and provide the most suitable microclimate (Hutchinson 1958; Spotila 1972), but
353 food items become more scarce there. This suggests the existence of a trade-off between the
354 selection of most external sectors (abundant prey, but unstable microclimate and high risk of
355 predation) and the deepest ones (lack of predators, suitable microclimate but scarce prey). We
356 propose that the habitat selection patterns may be explained by different risk-taking strategies
357 between age classes. For juveniles, accessing to relatively abundant food is necessary for growth
358 and development, therefore they choose external sectors, even if this may be risky.
359 In cave salamanders, adult survival is the most important determinant of population growth
360 rate (Lindstrom et al. 2010). This suggests that, after attaining sexual maturity, adults may trade off
361 food availability in favour of safe areas with stable micro-climate. A spatial distribution pattern
362 maximising survival after reaching sexual maturity would thus provide the highest fitness to
363 individuals. Overall, the age class found at the riskiest location may be not the one that experiences
364 the lowest risk of mortality there: segregation probably occurs because the mortality risk of
365 inhabiting the riskier habitat is offset by increased foraging requirements (Rochette and Grand
366 2004). The interplay between predation risk and resource availability has been observed multiple
367 times as a cause of intraspecific segregation. For instance, a similar strategy determines sexual
368 segregation in several polygynous ungulates, in which males need high-quality diet to be successful
369 in competition for mates, and thus select areas with superior food availability even if predation risk
370 may be high. Conversely, females maximise fitness by increasing offspring survival, and therefore
371 select the less risky areas, even if they have limited foraging quality (Main 2008). It should also
372 remarked that our study was not manipulative, and segregation among age classes may be caused by
373 other factors, including different optima for microclimate or additional unmeasured features.
374 The occurrence of intraspecific spatial segregation has lead to a flourishing number of
375 hypotheses (Ruckstuhl 2007). In many cases, potential causes of spatial segregation have been
376 proposed only after segregation was observed (inductive reasoning) (McIntire and Fajardo 2009).
15 377 Comprehensive analyses, particularly in the last years, allowed identifying the processes most likely
378 determining sexual segregation, and highlighted the importance of trade-offs between resources and
379 predation risk (Main and Coblentz 1996; see also Bon et al. 2001; Main 2008; Bjorneraas et al.
380 2012). Conversely, less attention has been devoted to age-class segregation. A hypothetico-
381 deductive reasoning, coupled with the comparison of explicit hypotheses, can allow inferring the
382 processes most likely determining distribution and segregation on the basis of distribution patterns
383 (McIntire and Fajardo 2009; Dochtermann and Jenkins 2011), thereby boosting our understanding
384 of age-class segregation. Our results are in agreement with studies highlighting the importance of
385 the trade-off between resources and predation risk: these trade-offs may be important also for age-
386 class segregation. Nevertheless, age-class segregation remains a much less studied topic, if
387 compared to sexual segregation: further analyses are required to better understand the costs and
388 benefits of different strategies of spatial distribution.
389
390 Aknowledgements
391
392 The comments of K. Ruckstuhl and one reviewer improved early versions of this manuscript. We
393 thank S. Salvidio for constructive discussions. E. Massa (Savonese Speleo Club) kindly provided
394 cave location data. We thank also M. Merazzi (Erba CAI Speleo Club). Data were collected during
395 the Herpetological field work of Natural Sciences students at the Università degli Studi di Milano.
396 We thank N. Santo, F. Basso, F. Belluardo, L. Borrelli, F. Barazzetta, E. Crenna, R. Gavazzi, L.
397 Limongi, E. Lunghi, L. Morotti, E. Orlando, F. Pantuso, M. Palombelli, A. Serioli, G. Soldà, S.
398 Virtuani, D. Zani and M. De Nicola for help during fieldwork. GFF was funded by a scholarship of
399 Univ. Milano-Bicocca. This paper is dedicated to Matteo.
400
401 References
402
16 403 Bates D and Maechler M (2010) lme4: linear mixed-effects models using S4 classes. R package
404 version 0.999375-37. www.r-project.org
405 Bjorneraas K, Herfindal I, Solberg EJ, Sther BE, van Moorter B and Rolandsen CM (2012) Habitat
406 quality influences population distribution, individual space use and functional responses in habitat
407 selection by a large herbivore. Oecologia 168: 231-243
408 Bologna MA and Salvidio S (2006) Speleomantes strinatii (Aellen, 1958). In: Sindaco R, Doria G,
409 Razzetti E and Bernini F (eds) Atlas of Italian Amphibians and Reptiles, Polistampa, Firenze, pp
410 258-261
411 Bon R, Rideau C, Villaret JC and Joachim J (2001) Segregation is not only a matter of sex in
412 Alpine ibex, Capra ibex ibex . Anim Behav 62: 495–504
413 Bonenfant C, Gaillard JM, Dray S, Loison A, Royer M and Chessel D (2007) Testing sexual
414 segregation and aggregation: Old ways are best. Ecology 88: 3202-3208
415 Bowyer RT, Stewart KM, Wolfe SA, Blundell GM, Lehmkuhl KL, Joy PJ, McDonough TJ and Kie
416 JG (2002) Assessing sexual segregation in deer. J Wildlife Manage 66: 536-544
417 Briggler JT and Prather JW (2006) Seasonal use and selection of caves by plethodontid salamanders
418 in a Karst area of Arkansas. Am Midl Nat 155: 136-148
419 Burnham KP and Anderson DR (1998) Model selection and inference. Springer Verlag, New York
420 Camp CD and Jensen JB (2007) Use of twilight zones of caves by plethodontid salamanders.
421 Copeia 2007: 594-604
422 Cimmaruta R, Forti G, Nascetti G and Bullini L (1999) Spatial distribution and competition in two
423 parapatric sibling species of European plethodontid salamanders. Ethol Ecol Evol 11: 383-398
424 Cransac N, Gerard JF, Maublanc ML and Pépin D (1998) An example of segregation between age
425 and sex classes only weakly related to habitat use in mouflon sheep ( Ovis gmelini ). J Zool 244: 371-
426 378
17 427 Crump ML and Scott NJ (1994) Visual encounter surveys. In: Heyer WR, Donnelly MA,
428 McDiarmid RW, Hayek LC and Foster MS (eds) Measuring and monitoring biological diversity:
429 standard methods for Amphibians. Smithsonian Institution Press, Washington, pp 84-92
430 Culver DC and Pipan T (2009) The biology of caves and other subterranean habitats. Oxford
431 University Press, Oxford
432 Darmon G, Calenge C, Loison A, Jullien J-M, Maillard D and Lopez J-F (2012) Spatial distribution
433 and habitat selection in coexisting species of mountain ungulates. Ecography: 44–53
434 Dochtermann NA and Jenkins SH (2011) Developing multiple hypotheses in behavioral ecology.
435 Behav Ecol Sociobiol 65: 37-45
436 Ficetola GF, Scali S, Denoël M, Montinaro G, Vukov TD, Zuffi MAL and Padoa-Schioppa E
437 (2010) Ecogeographical variation of body size in the newt Triturus carnifex : comparing the
438 hypotheses using an information-theoretic approach. Global Ecol Biogeogr 19: 485–495
439 Ficetola GF, Pennati R and Manenti R (2012) Do cave salamanders occur randomly in cavities? An
440 analysis with Hydromantes strinatii . Amphibia-Reptilia 33: 251-259
441 Field IC, Bradshaw CJA, Burton HR, Sumner MD and Hindell MA (2005) Resource partitioning
442 through oceanic segregation of foraging juvenile southern elephant seals ( Mirounga leonina ).
443 Oecologia 142: 127–135
444 Firth LB and Crowe TP (2010) Competition and habitat suitability: small-scale segregation
445 underpins large-scale coexistence of key species on temperate rocky shores. Oecologia 162: 163-
446 174
447 Formica VA, Gosner RA, Ramsay S and Tuttle EM (2004) Spatial dynamics of alternative
448 reproductive strategies: the role of neighbors. Ecology 85: 1125-1136
449 Galvan I (2004) Age-related spatial segregation of great cormorants in a roost. Waterbirds 27: 377-
450 381
451 Gautier P, Léna JP and Miaud C (2004) Responses to conspecific scent marks and the ontogeny of
452 territorial marking in immature terrestrial salamanders. Behav Ecol Sociobiol 55: 447-453
18 453 Hadfield JD (2010) MCMC methods for multi-response generalized linear mixed models: The
454 MCMCglmm R package. J Stat Softw 33: 1-22
455 Harvey V, Cote SD and Hammill MO (2008) The ecology of 3-D space use in a sexually dimorphic
456 mammal. Ecography 31: 371-380
457 Hillman SS, Whiters PC, Drewes RC and Hillyard SD (2009) Ecological and environmental
458 physiology of amphibians. Oxford University Press, New York
459 Hills N, Hose GC, Cantlay AJ and Murray BR (2008) Cave invertebrate assemblages differ between
460 native and exotic leaf litter. Austral Ecol 33: 271-277
461 Hutchinson VH (1958) The distribution and ecology of the cave salamander, Eurycea lucifuga . Ecol
462 Monogr 28: 1-20
463 Istock CA (1966) The evolution of complex life cycle phenomena: an ecological perspective.
464 Evolution 21: 592-605
465 Labée-Lund JH, Langeland A, Jonsson B and Ugedal O (1993) Spatial segregation by age and size
466 in Arctic Charr - a trade-off between feeding possibility and risk of predation. J Anim Ecol 62: 160-
467 168
468 Lanza B, Pastorelli C, Laghi P and Cimmaruta R (2006) A review of systematics, taxonomy,
469 genetics, biogeography and natural history of the genus Speleomantes Dubois, 1984 (Amphibia
470 Caudata Plethodontidae). Atti Mus Civ St Nat Trieste 52 (Suppl.): 5-135
471 Lindstrom J, Reeve R and Salvidio S (2010) Bayesian salamanders: analysing the demography of
472 an underground population of the European plethodontid Speleomantes strinatii with state-space
473 modelling. BMC ecology 10: 4
474 Lukacs PM, Thompson WL, Kendall WL, Gould WR, Doherty PF, Burnham KP and Anderson DR
475 (2007) Concerns regarding a call for pluralism of information theory and hypothesis testing. J Appl
476 Ecol 44: 456-460
477 Main MB (2008) Reconciling competing ecological explanations for sexual segregation in
478 ungulates. Ecology 89: 693–704
19 479 Main MB and Coblentz BE (1996) Sexual segregation in Rocky Mountain mule deer. J Wildlife
480 Manage 60: 497-507
481 Manenti R, Ficetola GF, Marieni A and De Bernardi F (2011) Caves as breeding sites for
482 Salamandra salamandra : habitat selection, larval development and conservation issues. North-
483 Western J Zool 7: 304-309
484 McIntire EJB and Fajardo A (2009) Beyond description: the active and effective way to infer
485 processes from spatial patterns. Ecology 90: 46-56
486 Moran NA (1994) Adaptation and constraint in the complex life cycle of animals. Annu Rev Ecol
487 Syst 25: 573-600
488 Novak T, Tkavc T, Kuntner M, Arnett AE, Delakorda SL, Perc M and Janzekovic F (2010) Niche
489 partitioning in orbweaving spiders Meta menardi and Metellina merianae (Tetragnathidae). Acta
490 Oecol 36: 522-529
491 Pastorelli C and Laghi P (2007) Predation of Speleomantes italicus (Amphbia: Caudata:
492 Plethofontidae) by Meta menardi (Arachnida: Araneae: Metidae). In: Bologna MA, Capula M,
493 Carpaneto GM, Luiselli L, Marangoni C and Venchi A (eds) Atti del 6° Congresso Nazionale
494 Societas Herpetologica Italica, Belvedere, Latina, pp 45-48
495 Pastorelli C, Laghi P and Scaravelli D (2005) Spacing of Speleomantes italicus (Dunn, 1923):
496 application of a geographic information system (G.I.S.) (Amphibia, Plethodontidae). In: Salvidio S,
497 Poggi R, Doria G and Pastorino MV (eds) Atti del Primo Convegno Nazionale Biologia dei
498 geotritoni europei Genere Speleomantes , Annali del Museo Civico di Storia naturale "G. Doria",
499 Genova, 97, pp 169-177
500 Pinheiro P and Bates D (2000) Mixed-effect models in S and S-PLUS. Springer, New York, 528 pp
501 Pinheiro P, Bates D, DebRoy S and Sarkar D (2010) Linear and nonlinear mixed effects models. R
502 package version 3.1-97. http://cran.r-project.org/web/packages/nlme
503 R Development Core Team (2010) R: A language and environment for statistical computing. R
504 Foundation for Statistical Computing, Vienna, www. r-project.org pp
20 505 Rochette R and Grand TC (2004) Mechanisms of species coexistence: a field test of theoretical
506 models using intertidal snails. Oikos 105: 512-524
507 Ruckstuhl KE (2007) Sexual segregation in vertebrates: proximate and ultimate causes. Integr
508 Comp Biol 47: 245–257
509 Ruckstuhl KE and Festa-Bianchet M (2001) Group choice by subadult bighorn rams: Trade-offs
510 between foraging effciency and predator avoidance. Ethology 107: 161-172
511 Salvidio S (1993) Life history of the European plethodontid salamander Speleomantes ambrosii
512 (Amphibia, Caudata). Herpetol J 3: 55-59
513 Salvidio S and Pastorino MV (2002) Spatial segregation in the European plethodontid
514 Speleomantes strinatii in relation to age and sex. Amphibia-Reptilia 23: 505-510
515 Salvidio S, Lattes A, Tavano M and Melodia F (1994) Ecology of a Speleomantes ambrosii
516 population inhabiting an artificial tunnel. Amphibia-Reptilia 15: 35-45
517 Schneider K, Christman MC and Fagan WF (2011) The influence of resource subsidies on cave
518 invertebrates: results from an ecosystem-level manipulation experiment. Ecology 92: 765-776
519 Singh NJ, Bonenfant C, Yoccoz NG and Cote SD (2010) Sexual segregation in Eurasian wild
520 sheep. Behav Ecol 21: 410-418
521 Smithers P (2005) The diet of the cave spider Meta menardi (Latreille 1804) (Araneae,
522 Tetragnathidae). Journal of Arachnology 33: 243-246
523 Spiegelhalter DJ, Best NG, Carlin BR and van der Linde A (2002) Bayesian measures of model
524 complexity and fit. J R Stat Soc Ser B-Stat Methodol 64: 583-616
525 Spiegelhalter D, Thomas A, Best N and Lunn D (2008) WinBUGS 1.4.3. Imperial College and
526 MRC, UK, http://www.mrc-bsu.cam.ac.uk/bugs/winbugs/contents.shtml pp
527 Spotila JR (1972) Role of temperature and water in the ecology of lungless salamanders. Ecol
528 Monogr 42: 95-125
529 Stephens PA, Buskirk SW, Hayward GD and Del Rio CM (2007) A call for statistical pluralism
530 answered. J Appl Ecol 44: 461-463
21 531 Symonds MRE and Moussalli A (2011) A brief guide to model selection, multimodel inference and
532 model averaging in behavioural ecology using Akaike’s information criterion. Behav Ecol
533 Sociobiol 65: 13-21
534 van Toor ML, Jaberg C and Safi K (2011) Integrating sex-specific habitat use for conservation
535 using habitat suitability models. Anim Conserv 14: 512-520
536 Vignoli L, Caldera F and Bologna MA (2006) Trophic niche of cave populations of Speleomantes
537 italicus . Journal of Natural History 40: 1841-1850
538 Vignoli L, Caldera F and Bologna MA (2008) Spatial niche of the Italian cave salamander,
539 Speleomantes italicus (Dunn, 1923) (Plethodontidae, Amphibia), in a subterranean system of
540 Central Italy. Ital J Zool 75: 59-65
541 Wilbur HM (1980) Complex life cycles. Annu Rev Ecol Syst 11: 67-93
542
543
22 544 Fig. 1 . Distribution patterns predicted by different hypotheses proposed to explain spatial 545 segregation. The social segregation hypothesis (a) predicts a negative relationship between the 546 distribution of the different classes but, (b) when this relationship is taken into account, similar 547 responses to environmental gradients. The habitat segregation hypothesis predicts (c) a non- 548 negative relationship between the abundance of different classes, but (d) different responses to 549 environmental gradients. Different classes are here exemplified by age classes [juveniles (white 550 lines) vs. adults (black lines)], but sex, morph or other classes can be similarly considered. 551 Social segregation hypothesis a b Abundance Abundance of juveniles Environmental features Abundance of adults
Habitat segregation hypothesis
c d Adults
Juveniles Abundance Abundance of juveniles
Abundance of adults Environmental features
23 552 Fig. 2 . Box-plot representing the distance of juvenile and adult salamanders from the cave entrance. 553 Bold lines represent the median of groups. The circle on top right for adults represents an outlier. 554
50
40
30
20
10 Distance from cave entrance (m)
0
555 Juveniles Adults
24 556
557 Fig. 3 . Relationship between the abundance of spiders and the abundance of (a) juvenile cave 558 salamanders; (b) adult cave salamanders. Error bars are standard errors; the scales of the vertical 559 axes differ between the 560 panels 2.5 a) b) 1.5
2.0
1.0 1.5
1.0 0.5 0.5 Abundance of juveniles Abundance of adults
0.0 0.0 0 1-3 4-6 >6 0 1-3 4-6 >6 561 Abundance of spiders Abundance of spiders
25 562 Table 1 Results of generalized linear mixed models describing the abundance of juveniles and
563 adults on the basis of the social (a, e), habitat (b, f) and combined hypotheses (c, g). We also report
564 the results of Bayesian Markov Chain Monte Carlo models (MCMCglmm) with lowest DIC (d, h;
565 see Table 3 for DIC values). Significant values are in bold.
566
Dependent: abundance of juveniles Dependent: abundance of adults 2 2 Independent B χ 1 P AIC Independent B χ 1 P AIC
Social models a) N adults 0.12 1.86 0.172 97.54 e) N juveniles 0.13 1.17 0.280 99.51
Habitat models b) Temperature 0.02 0.017 0.896 88.86 f) Temperature -0.31 6.23 0.013 88.25 Humidity -0.04 2.65 0.104 Humidity -0.04 1.88 0.170 Illuminance -0.38 0.44 0.506 Illuminance -2.14 3.39 0.066 Spider abundance 0.58 12.03 0.002 Spider abundance -0.45 4.51 0.034
Combined models c) N adults 0.47 9.31 0.002 81.55 g) N juveniles 0.40 6.01 0.014 84.24 Temperature 0.27 3.78 0.052 Temperature -0.30 6.74 0.009 Humidity -0.02 0.84 0.360 Humidity -0.03 1.07 0.300 Illuminance -0.08 0.02 0.887 Illuminance -2.67 4.25 0.039 Spider abundance 0.83 19.32 <0.001 Spider abundance -0.70 8.01 0.005
d) best MCMCglmm MCMC P h) best MCMCglmm MCMC P
N adults 0.77 0.010 N juveniles 0.47 0.045 Temperature 0.41 0.114 Temperature -0.36 0.012 Humidity -0.02 0.619 Humidity -0.01 0.601 Illuminance -0.82 0.615 Illuminance -3.20 0.043 Spider abundance 1.26 0.002 Spider abundance -0.87 0.007 567
26 568 Table 2 a) Comparison of Akaike Information Criterion (AIC) values of generalized linear mixed
569 models describing the abundance of juveniles and adults on the basis of the social, ecological and
570 combined hypotheses, considering only sectors in which at least one salamander was present (either
571 adults or juveniles). b) Comparison of AIC values of models considering only sectors in which at
572 least two salamanders (either adults or juveniles) were present. c) Coefficients and significance of
573 individual variables for the model with lowest AIC in Table 2a. Significant values are in bold.
574
Dependent: Dependent: abundance of juveniles abundance of adults a) Comparison among models: AIC AIC sectors with at least one salamander
Social models 58.88 49.79 Combined models 50.76 41.53 Habitat models 49.92 40.47
b) Comparison among models: sectors with at least one salamander
Social models 30.93 30.12 Combined models 31.87 26.55 Habitat models 29.99 24.55
c) Coefficients of independent variables 2 2 B χ 1 P B χ 1 P Temperature 0.02 0.02 0.896 -0.33 8.88 0.003 Humidity -0.01 0.08 0.775 -0.01 0.41 0.525 Illuminance 1.50 1.29 0.257 -0.30 0.03 0.868 Spider abundance 0.58 7.65 0.006 -0.49 4.45 0.035 575
27 576 Table 3 Comparison of Deviance Information Criterion (DIC) between MCMCglmms built using a
577 zero-inflated Poisson family, and MCMCglmms built using a Poisson family. The independent
578 variables included in the social, in the ecological and in the combined models are listed in Table 1.
579
Dependent: abundance of juveniles Dependent: abundance of adults
Zero-inflated Poisson Zero-inflated Poisson
DIC DIC DIC DIC
Social model 112.8 97.9 145.1 133.5
Habitat model 106.4 97.9 139.9 133.3
Combined model 104.8 93.3 139.6 131.5
580
581
582
28